Depth-Based Object Re-Identification

4. The method of claim 1, wherein the frame-level confidence is determined using a convolutional neural network.

5. The method of claim 4, wherein the convolutional neural network includes a plurality of layers, and wherein different layers of the plurality of layers have different learning rates.

6. The method of claim 1, wherein the video includes depth video captured by a depth camera.

7. The method of claim 6, wherein for each of the plurality of frames of the depth video captured by the depth camera, the frame-level confidence is determined using a convolutional neural network that includes a first subset of layers derived from a color-based model previously-trained to classify a color frame, wherein the first subset of layers has a first learning rate, wherein the convolutional neural network includes a second subset of layers having a second learning rate that is higher than the first learning rate, and wherein the first subset of layers are closer to an input layer of the convolutional neural network than the second subset of layers.

8. The method of claim 1, wherein weighting of the weighted determined frame-level confidence increases with increasing assessed quality of the frame.

10. The storage machine of claim 9, wherein the instructions further comprise instructions to assess a current integrated confidence based on the frame-level confidence and a previous integrated confidence assessed for a previous frame of the video, and wherein the overall confidence is assessed based on the current integrated confidence in addition to the weighted determined frame-level confidence.

11. The storage machine of claim 9, wherein the instructions further comprise instructions to, for a plurality of different previously-recognized objects, assess the overall confidence that the previously-recognized obj ect is present in the video based on weighted determined frame-level confidences for each of the plurality of frames of the video corresponding to the previously-recognized object.

12. The storage machine of claim 9, wherein the frame-level confidence is determined using a convolutional neural network.

13. The storage machine of claim 12, wherein the convolutional neural network includes a plurality of layers, and wherein different layers of the plurality of layers have different learning rates.

14. The storage machine of claim 9, wherein the video includes depth video captured by a depth camera.

15. The storage machine of claim 14, wherein for each of the plurality of frames of the depth video captured by the depth camera, the frame-level confidence is determined using a convolutional neural network that includes a first subset of layers derived from a color-based model previously-trained to classify a color frame, wherein the first subset of layers has a first learning rate, wherein the convolutional neural network includes a second subset of layers having a second learning rate that is higher than the first learning rate, and wherein the first subset of layers are closer to an input layer of the convolutional neural network than the second subset of layers.

17. The method of claim 16, wherein the frame-level confidence is determined using a convolutional neural network.

18. The method of claim 17, wherein the convolutional neural network includes a plurality of layers, and wherein different layers of the plurality of layers have different learning rates.

19. The method of claim 16, wherein the video includes depth video captured by a depth camera.

20. The method of claim 19 wherein for each of the plurality of time-sequential frames of the depth video captured by the depth camera, the frame-level confidence is determined using a convolutional neural network that includes a first subset of layers derived from a color-based model previously-trained to classify a color frame, wherein the first subset of layers has a first learning rate, wherein the convolutional neural network includes a second subset of layers having a second learning rate that is higher than the first learning rate, and wherein the first subset of layers are closer to an input layer of the convolutional neural network than the second subset of layers.